Initial Benchmarks Of The Intel Downfall Mitigation Performance Impact

First up are some benchmarks of the old (390) vs. new (3a5) microcode on a Intel Xeon Platinum 8380 2P server, which was the flagship Ice Lake server processors up until the debut of Xeon Scalable Sapphire Rapids in January.

Somewhat ironic is that many of the workloads that were quickly found to be negatively impacted by the CPU microcode mitigation for Downfall were Intel's own software packages. Though not too surprising considering all of the AVX2/AVX-512 optimizations carried out by Intel software engineers in aiming to maximize their software's performance on Intel processors. Intel's Embree ray-tracing library that is also used by other software was running slower with the new Ice Lake microcode.

Intel's Open Volume Kernel Library (OpenVKL) was about 6% slower with the new CPU microcode.

When turning to the OSPRay engine though is where some big hits were appearing to the performance as a result of the mitigated microcode for Downfall.

This also carried over into using the OSPRay Studio software suite as a whole was leading to longer render times due to the Downfall mitigation.

These are real-world software packages seeing slower performance as a result of the new CPU microcode. It's not just synthetic tests but software packages as part of Intel's oneAPI software suite that Intel widely promotes. Presumably these software packages are already following Intel's own software best practices and guidance, but we'll see if any new releases try to lessen the blow of the Downfall mitigation or pursue other optimizations to offset the losses.